CFD-aided design improves FCC performance

D. Barthod, M. Del Pozo, C. Mirgain
Total Raffinage Distribution SA
Harfleur, France

Total Raffinage Distribution SA and Institut Français du Pétrole (IFP) successfully modified commercially available computational fluid dynamics (CFD) code to more accurately represent fluid-solid flow in a fluid catalytic cracking unit (FCCU).

Using this modified code in 1994, Total simulated the nonuniform catalyst flow in Total's refinery near Dunkirk, France. From this simulation, they were able to correct the catalyst flow by adding several steam jets to the J-bend.

The customized CFD simulations have been used to accelerate research and development (R&D) and develop new reactor-riser configurations to improve catalyst and feed contacting. CFD has also proven to be a cost-effective means of analyzing the hydrodynamics in the FCCU to troubleshoot and recommend solutions to improve the operations.

This article summarizes work done to improve the catalyst and feed contacting using CFD simulation at the FCCU at Total's refinery near Dunkirk. Total installed a new counter-current feed-injection (CCFI) system, a new riser-termination device, and new stripper internals to the existing Dunkirk reactor.

FCC process at Total

Total operates six FCCUs around the world with capacities ranging from 20,000 to 50,000 b/d. As an operator, Total has a direct economic interest in improving FCCU performance.

Over the past 20 years, Total has also conducted intensive research in fluid-catalytic cracking (FCC) technology resulting in innovative developments. In the early 1980s, Total developed the original two-regenerator, one-reactor (R2R) resid-catalytic cracking process which has become the most widely used resid FCC technology.

IFP and Stone & Webster Engineering Corp. (S&W) now license the R2R process with ongoing R&D support by Total.

One of the most important aspects of the R2R technology is the catalyst and feed contacting. This contact is a critical process step when injecting residual feeds that cannot fully vaporize. The revolutionary reactor-riser design and feed injectors that were developed as part of the R2R technology almost 20 years ago have been successfully applied to over 2 million b/d of FCC capacity. More recent work has focused on the next challenge of ultra-short contact time catalytic cracking or Ultra Selective Cracking (USC).

Early pilot work over 6 years ago identified a potential to improve product selectivity by precisely controlling the catalyst and feed contacting and residence time. Translating the results in the lab into a full commercial-scale plant presented several challenges.

To achieve short contact time within several hundred milliseconds, it is necessary to mix the feed and catalyst extremely rapidly and use the contact time as effectively as possible.

Traditional up-flow reactor risers are known to promote a classic core-annulus flow pattern just above the feed-injection zone. The vapor tends to concentrate in the center of the riser, and the catalyst tends to collect at the wall. The objective is to retain or enhance the catalyst and feed mixing and maintain the mixture without segregation.

Total and IFP studied the problem using cold-flow units and CFD modeling to complement pilot plant work to define yield enhancements using USC. The use of CFD code to simulate two-phase, fluid-solid flow is by no means well proven nor widely demonstrated with commercially available software. Small particle interactions are complex with a tendency to form clusters that are difficult to model accurately.

The opportunities to accelerate R&D work and enhance technical services capabilities to support its refineries has led Total to customize commercially available CFD code and improve on the fluid-solid physics packages. Through collaboration with IFP, modified CFD code has been implemented and validated using large-scale cold flow data as well as commercial data. The resulting modified code is a useful tool for R&D and refinery technical support.

The cold flow, pilot plant, and CFD modeling work led to a large demonstration test loop using CCFI into a down-flow reactor. Before building this demonstration unit, however, the use of CCFI as a means of intensifying the catalyst and feed mixing zone was commercialized in Total's FCCU in the Dunkirk refinery.

A sketch of the revamped unit is shown in Fig. 1 [60,377 bytes]. The unit originally started up in 1982 as an Exxon Flexicracker. Three modifications were made to the reaction system in October 1994: a new CCFI system, a new riser-termination device, and new stripper internals comprising structured packing.

The Dunkirk revamp was very successful in terms of increased product yield and qualities. A 6 wt % increase in conversion was observed at similar riser outlet temperatures and equivalent feed and catalyst quality.

A classical, global approach for post-revamp performance evaluation was applied to evaluate the change in unit performance before and after the revamp. Total uses a kinetic yield model developed in conjunction with IFP and S&W which can be used to track the actual unit performance as compared to the model to detect fundamental changes in operation.

The LPG and light gasoline selectivities increased, whereas the slurry yield dropped significantly (Fig. 2) [53,363 bytes]. Most of these trends were expected considering the technology changes made to the unit. The sharp increase in conversion and the decline in slurry oil yield, however, were not easily understood in relation to the process changes.

Evaluation of catalyst distribution

Numerous tests were conducted to study the performance of the new reaction system. These tests included reactor-mix sampling (RMS), gamma scans of the riser, and radial and axial temperature profile measurements. The Dunkirk riser was equipped with sampling ports located at various heights above the feed-injection zone to facilitate testing and unit validation.

In one test, the riser-temperature profile was measured using special probes inserted into the riser-sampling ports. Surprisingly, it was observed that the temperature profiles a few meters above the feed injection were not symmetrical.

The riser radial-temperature profile measured is shown in Fig. 3 [109,969 bytes]. The temperature difference between the hottest and coldest recorded spots was as high as 25° C. The maximum temperature zone was found to be at 555° C. for a riser outlet temperature of 515° C. and calculated reaction mix temperature of 555° C. for the test conditions employed. Clearly, the mixing of hot catalyst and the feed was nonuniform and likely reducing reactor performance.

The first concern was that the feed injectors were not distributing the feed uniformly. Feed and steam flow rates to each injector were thoroughly checked and found to be uniform. This left the catalyst flow and distribution as the likely cause. By comparing temperature profiles at various heights, the asymmetric temperature profiles were found to be somewhat related to the axis of the J-bend (Fig. 3). The riser zone located above the outside of the turn was always much hotter than the riser zone above the inside of the turn.

To confirm this observation and evaluate the penalty of operating in this situation, a new test was designed. Together with the measurement of temperature profiles a few meters above the feed injection, RMS tests were performed in the vapor line to evaluate the yields and selectivities.

A base case was done with all the injectors in operation. Then feed injectors 7 and 8 (Fig. 3) were turned off to match the hypothesized imbalance in catalyst flow.

The temperature profile became symmetrical and the temperature difference between the hottest and coldest spots dropped from 25 to 11° C. More importantly, selectivities were significantly improved as shown in Table 1 [20,187 bytes] . The imbalance in catalyst flow must have been significant if turning off two feed injectors actually improved the unit performance.

An explanation for the observed performance was proposed: The distance between the J-bend turn and feed injection may have been too short to obtain a uniform distribution of catalyst at the feed nozzles level.

By nature, the gas-solid mixture density profile in the long inclined line leading to the bottom of the riser is not uniform as observed in cold flow modeling and scans of the J-bend. Although gas injection along the J-bend and at the turn to vertical is supposed to promote even catalyst distribution, the nonuniform flow in the bend cannot be corrected at the turn with the steam-injection points provided.

CFD modeling

Continued operation of the unit without two feed injectors was out of the question. The R&D team was called upon to find a solution. The complex catalyst and gas flow patterns in the J-bend and riser were a formidable modeling challenge, but previous success with CFD modeling suggested a solution could be found using a scaled simulation of the Dunkirk reaction system.

The objective of the CFD study was to understand the different phenomena generating the segregation in the catalyst flow, and then propose a design which would improve the catalyst distribution just below the feed injectors.

CFD modeling was done using the exact J-bend geometry but on a 1:3 reduced scale to limit calculation times. A snapshot of the volume fraction of catalyst in the J-bend after a few seconds of simulation is shown in Fig. 4a [73,096 bytes].

The CFD model confirmed the team's suspicions (and visual observations in cold flow models) that catalyst and gas were not distributed uniformly in the J-bend.

After the first turn at the bottom of the J-bend, gas and solids tend to separate by inertial effects; gravity forces are more active on the dense catalyst particles. This mechanism creates bubbles that move up along the top wall of the inclined line and disturb the feed injection area downstream of the J-bend.

The J-bend is configured with numerous steam-aeration injection points to counter-balance the gravity effects and slow the catalyst deaeration near the bottom wall. However, the aeration rates are not high enough to allow a uniform flow of catalyst in this line.

The results of a more complete time-averaged simulation are shown in Fig. 4b. As before, the distribution of catalyst is shown to be poor after the first turn (bottom of the J-bend). These simulations were confirmed by gamma ray measurements on the commercial unit where zones of high and low density were observed.

The disturbance created in the J-bend propagates through the riser to the feed injection zone approximately 3 m above the J-bend.

In Fig. 5 [37,640 bytes], the simulated time-averaged radial profile of void fraction (blue curve) is shown at the feed-injection level for the existing configuration. The void fraction in the riser zone above the outside of the turn (riser bottom) is 40% lower than the cross section average-that is, the catalyst concentration is much higher in this zone.

To create a better distribution of the catalyst in the feed-injection zone, various modifications to the existing configuration were considered. Several modifications to the steam-injection rate and distribution were examined as possible solutions. Based on CFD simulations, it appeared that an efficient and low-cost solution was to modify the pattern of steam injection in the riser prior to feed injection as a means of improving the catalyst flow profile.

New steam-injection points were designed to optimize the correcting effect for the minimum amount of steam injected. These redistribution steam-injection points were implemented on the outside of the bend, at the bottom of the riser between the turn and the feed injection.

The result of the modified design is shown in Fig. 5. The simulated void fraction profile at the feed-injection level is much more uniform after adding the redistribution steam-injection points (red curve in Fig. 5). Using a steam flow rate of only 1 ton/hr (less than 0.5 wt % relative to feed) the density profile variations are reduced to less than 10% throughout the riser cross-section. During the next unit turnaround, these modifications will be installed.

Acknowledgments

The authors would like to thank D. Chombart and M.A. Senegas for their help preparing the unit tests. They also thank Y. Buvot and M. Lebouteiller for carrying the on-site experiments.

The Authors